NotI

NotI is an AI-first note-taking desktop app.

Note + AI = NotI: you capture ideas quickly, then use built-in AI to structure, rewrite, summarize, or expand them without leaving your notes workflow.

At a high level, NotI combines:

• local-first note and folder management
• speech-to-text (Whisper) for voice capture
• LLM-powered writing assistance (local models or API provider)
• a native desktop experience built with Wails (Go backend + TypeScript frontend)

• Xcode Command Line Tools
  • xcode-select --install
• Go 1.24+
• Bun (for frontend dependencies and scripts)
• Wails CLI v2
  • go install github.com/wailsapp/wails/v2/cmd/wails@latest

git clone <your-fork-or-repo-url>
cd noti
bun install --cwd frontend

./scripts/test.sh
./scripts/build.sh debug
./scripts/build.sh production

./scripts/build.sh debug generates build/bin/noti.app.

./scripts/build.sh production generates build/bin/noti.app and packages it into build/bin/noti.dmg.

Current build/test scripts assume a local whisper.cpp checkout and macOS CGO flags are available. If your local path differs, set WHISPER_PATH before building/testing.

WHISPER_PATH=/absolute/path/to/whisper.cpp ./scripts/test.sh
WHISPER_PATH=/absolute/path/to/whisper.cpp ./scripts/build.sh production

If build/link errors mention libwhisper/ggml ABI or deployment target mismatches, rebuild the local whisper artifacts:

./scripts/rebuild-whisper.sh

You can pass a custom whisper checkout path:

./scripts/rebuild-whisper.sh /absolute/path/to/whisper.cpp

Pushing a tag matching v* (for example v1.2.3) triggers GitHub Actions to build unsigned DMGs for both macOS architectures and attach them to the GitHub release:

• noti-macos-intel.dmg
• noti-macos-apple-silicon.dmg

NotI is a Wails desktop app with a Go backend and a TypeScript frontend.

• frontend/
  • Vite + TypeScript UI + Tailwind CSS
  • stateful note editor, sidebar, settings, prompts, and runtime events
• main.go
  • Wails app bootstrap, menu setup, logging, and production error reporting wiring
• app.go
  • Wails-bound application methods exposed to the frontend
  • orchestration across notes, folders, config, STT, LLM, prompts, and audio
  • on first run, STT/LLM initialization (and model downloads) is deferred until the Getting Started setup is saved
• internal/service/
  • service layer for business logic and lifecycle management
  • includes STTManager, LLMManager, AudioManager, NoteService, FolderService, and PromptService
• internal/stt/whisper/
  • real-time Whisper transcription pipeline with VAD, chunking, assembly, and event emission
• internal/llm/
  • provider abstraction for local (llama-server) and API-based LLMs
  • model/binary download and runtime health management
• internal/infrastructure/downloader/
  • model and binary download registry + progress callbacks

Storage defaults:

• Notes: ~/Documents/noti/notes
• Folder structure metadata: ~/Documents/noti/notes/structure.json
• App config/models/binaries: user config directory under Noti (for example on macOS: ~/Library/Application Support/Noti)

First-run behavior:

• config.json is created from the embedded template on first launch.
• The Getting Started screen is shown and model downloads do not start automatically.
• Model initialization/download begins only after the user clicks save in Getting Started.

macOS-first development environment:

• macOS with microphone permissions enabled for the app/terminal during development
• Go 1.24+
• Bun 1.x+
• Wails v2 CLI
• CGO-enabled Go toolchain
• Native dependencies used by this project:
  • whisper.cpp (Go bindings)
  • PortAudio (development/build dependency via Homebrew)
  • Apple frameworks used in current scripts (Accelerate, Foundation, Metal)

Release packaging note:

• macOS production builds bundle libportaudio.2.dylib inside noti.app, so end users do not need to install PortAudio separately.

Optional but recommended:

• Local LLM runtime support for llama-server model execution
• API endpoint + API key if using API provider mode

NotI is designed to be local-first:

• Notes are stored as local files on your machine.
• STT and local LLM flows can run entirely on-device.
• Model files and binaries are downloaded to local application directories.

When external services are used:

• If you configure an API-based LLM provider, prompts/content sent to that provider leave your machine.
• API credentials are stored in local app configuration.

Good practices:

• Use local provider mode for sensitive notes when possible.
• Treat API keys like secrets and rotate them if exposed.
• Exclude personal note directories and local config from backups/shares if needed.